Re: RFC: [PATCH] Fix setpriority calling __task_priority() for processes instead of threads.

[Samuel Thibault]

Svante Signell, on Thu 01 Sep 2016 12:44:04 +0200, wrote:
> On Wed, 2016-08-31 at 14:16 +0200, Samuel Thibault wrote:
> > Svante Signell, on Wed 31 Aug 2016 14:10:22 +0200, wrote:
> > > 
> > > 
> > > I would expect values [-20,19] to be OK converted to [5,44] with
> > > #define NICE_TO_MACH_PRIORITY(nice) ((nice) + 25) from 
> > > hurd/hurd/resource.h
> > > and 
> > > #define invalid_pri(pri) (((pri) < 0) || ((pri) >= NRQS)), where NRQS = 50
> > > from kern/sched.h.
> > > to work.
> > Yes, but see the code changing thread priorities (since that's what is
> > posing problem)
> 
> Digging a little further shows that tasks and threads are created with
> BASEPRI_USER=25 (NICE_TO_MACH_PRIORITY(0)=25). On the other hand kernel 
> threads
> has a max priority of BASEPRI_SYSTEM=6 (NICE_TO_MACH_PRIORITY(-19)=6). 
> (is -20 not possible as for Linux??)

These are just defaults, they could be changed to do -20, but POSIX
doesn't require that, so we don't need to care.

> In the code kern/task.c:task_priority() we have
> if (change_threads) {
> ...
> if (thread_priority(thread, priority, TRUE)!= KERN_SUCCESS) ret = 
> KERN_FAILURE;
> and in kern/thread.c:thread_priority():
> if (priority < thread->max_priority) {ret = KERN_FAILURE;}
> irrespective of being root or not.

So that's the issue.

> Should code be added to check for root(how)
> in task-priority() and call kern/thread.c:thread_max_priority() for all 
> threads
> in the task? Or are tasks created by root different in gnumach?

They aren't created differently. For mach, being root means having
a right on the privileged host port. Check how it is done in glibc's
sysdeps/mach/hurd/mlock.c and gnumach's vm/vm_user.c which checks for
ip_kotype(port) == IKOT_HOST_PRIV.

So the idea could be that we add to gnumach an RPC which additionally
takes a host port, and can thus check whether it's privileged or not.

Samuel